Ott Velsberg, Estonia’s fresh-faced, 28-year-old chief data officer, is on a mission put AI into every part of the country's public services, from healthcare to education and job centres.
“The aim is to make government more proactive and responsive to people’s life-events,” says Velsberg. Instead of citizens having to apply for things like driver's licences and school places, he envisions a system where public bodies can anticipate and preemptively respond to the needs people have at different stages of their life.
We’re not telling people what to do. That might happen in China but not in Estonia, or in Europe as a whole.
Take enrolling children in school. There is no reason why parents should have to apply for a place for their children, says Velsberg. You can calculate a child’s school needs based on birth data obtained from hospitals. Estonia is due to automated the process by the end of this year, says Velsberg adding that the programme doesn’t even really need AI to run.
But AI comes into play for more complex problems, such as calculating when people are called in to see a doctor. Estonia worked with Microsoft and the World Bank to develop a solution that would scan healthcare records to help doctors decide when patients need to be called in for medical checkups.
“For example, a patient with a disability and also suffering from heart disease may need to visit the doctor much more frequently, maybe every month, while someone with just diabetes and no other problems needs to be seen every six months,” explains Velsberg. The AI helps doctors manage their patient lists and time more effectively.
Another, slightly less developed plan is to create personalised, sports advice for teens, which would take into account puberty and recommend an exercise regime suited to their particular developmental stage.
I ask Velsberg whether he thinks teenagers would take advice from a government-run AI bot. He admits he isn’t sure.
“How we give information out is still up for debate,” he says. He’s quick to add that the AI would not dictate to people.
“We’re not telling people what to do, we are giving options to consider. You can still make up your own mind. The government can never go to the individual level of telling you what to do. That might happen in China but not in Estonia, or in Europe as a whole.”
Velsberg seems, to be honest, a little shaken by THAT article in Wired earlier this year, which outlined his plan to develop a “robot judge” for Estonia’s Ministry of Justice, where an AI programme would review legal documents and adjudicate in small claims disputes. The plan gained a huge amount of attention (and concern) in the media — Google Velsberg’s name and you get pages and pages of news stories about it.
He says he realises now that outside of Estonia there is considerably more concern about AI. In Estonia, he maintains, it is not a big deal.
“I haven’t seen any negative reaction to the use of AI [in Estonia]. Everyone is used to everything being digital. The government is being extremely transparent and everyone can see what data is being collected.”
Estonia is the world’s most advanced e-government. The small Baltic nation has had digital ID for its 1.3m citizens since 2002, online voting since 2005, and put digitised health records on the blockchain more than a decade ago. Using AI to automate services is a logical next step. Velsberg was hired last August with a specific remit to introduce AI into various ministries.
Velsberg says there are already some 16 government processes that have been automated using AI, and the healthcare bookings and sports advice are just two of a long list of projects he wants to test out.
The government has a €10m fund to help get innovative projects, including AI-related ones, off the ground — €5m for very experimental ideas that still need to be tested and the rest for more established projects that need to be scaled up.
Projects include everything from using natural language processing to replace court stenographers, to using machine learning to calculate the most efficient routes for the icebreakers that keep shipping lines clear in winter, a project that Velsberg says could save €1m.
Around 72% of people who were suggested a job by the AI system were still employed six months later, compared with 58% of those advised by a human.
There have been some early successes. Job centres have been equipped with AI that can parse people’s work history to try to match them with the job they are best suited for. The AI system appears so far to have a better hit rate than its human counterparts: around 72% of people who were suggested a job by the AI system were still employed six months later, compared with 58% of those advised by a human.
The government has also saved €665,000 on farming subsidies by using AI and satellite images to determine whether farmers really have mown their hay fields or not to qualify for a payout. Previously inspectors would have had to visit farms in person, and were able to cover less than 10% of the country's farmland.
The system is more than 85% accurate, Velsberg says. But changes of light can still confuse the programme, and a rain-flatten field may appear mown. He is keen to stress that machine learning like this will never be able to entirely replace humans.
“We are not getting rid of people, but enhancing what they can do,” he says.
As if to underscore the point, Estonia has recently opted not to give artificial intelligence separate legal status. Since 2016 a special task force had been considering this so-called “kratt law” (kratts are creatures in Estonian mythology, made of inanimate household objects that come to life — and another handy metaphor for AI) but had come to the conclusion that this was unnecessary.
“Our legal discussions came to a conclusion that at this point we do not need "kratt law",” says Velsberg. “Kratts remain a tool that humans use, thus humans are responsible.”